Latency-aware computation offloading and DQN-based resource allocation approaches in SDN-enabled MEC
Latency-aware computation offloading and DQN-based resource allocation approaches in SDN-enabled MEC
复制标题
支持 SDN 的 MEC 中的延迟感知计算卸载和基于 DQN 的资源分配方法
DOI：
10.1016/j.adhoc.2022.102950
复制
发表时间：
2022-07
期刊：
影响因子：
4.8
通讯作者：
Youlong Luo
中科院分区：
文献类型：
--
作者：
Tianyu Du;Chunlin Li;Youlong Luo
提出的移动边缘计算可以将移动应用中的计算任务转移到附近的边缘设备，有效降低本地服务器的处理压力，避免回程和核心网的延迟，从而更好地解决云计算无法有效处理资源分配的问题。然而，由于通信过程中无线环境的复杂性，导致通过无线链路上传到MEC端的流量和丢包数量增加，容易导致计算任务增加，无法保证较低的系统总能耗和较短的总时延。针对移动设备无法及时处理大量计算密集型任务的问题，本文提出了一种基于sdn的MEC环境下的任务卸载优化方案。基于李雅普诺夫优化对计算卸载问题进行建模，分析了卸载延迟与卸载增益的关系。为了保证应用需求，最小化能耗和时延，更好地满足用户QoS要求，本文提出了一种基于深度强化学习的资源分配策略。该策略设计了一种基于dqn的资源分配算法，在有限的计算资源和计算任务的延迟约束下，在移动边缘计算环境中部署联合最优卸载决策和资源分配方案。实验结果表明，所提出的任务卸载策略可以降低整体延迟；所提出的资源分配策略可以降低系统的总能耗和总时延，提高任务的成功率。
The proposed mobile edge computing can transfer the computing tasks in mobile applications to the nearby edge devices, effectively reducing the processing pressure of local servers and avoiding delays in backhaul and core networks, thus better solving the problem that cloud computing cannot effectively handle resource allocation. However, the complexity of the wireless environment during the communication process leads to the fact that the computing tasks are easily caused by the increase in the number of traffic and packet loss when they are uploaded to the MEC side through the wireless link, which cannot guarantee a lower total system energy consumption and shorter total delay. To address the problem that mobile devices cannot handle many computationally intensive tasks in a timely manner, this paper proposes a task offloading optimization scheme for SDN-enabled MEC environments. Modeling the computation offloading problem based on Lyapunov optimization, and then analyzes the offloading delay with respect to the offloading gain. In order to guarantee application requirements, minimize energy consumption and latency, and better satisfy user QoS requests, this paper proposes a resource allocation strategy based on deep reinforcement learning. The strategy designs a DQN-based resource allocation algorithm to deploy a joint optimal offloading decision and resource allocation scheme in a mobile edge computing environment under the limited computational resources and the latency constraints of the computational tasks. Based on the experimental results, it is shown that the proposed task offloading strategy can reduce the overall latency; the proposed resource allocation strategy can reduce the total energy consumption and total latency of the system and improve the successful execution rate of tasks.
登录
查看更多内容
DOI：
10.1007/s10586-020-03226-8
发表时间：
2021-01
期刊：
Cluster Computing
影响因子：
--
作者：
Yashwant Singh Patel;M. Reddy;R. Misra
通讯作者：
Yashwant Singh Patel;M. Reddy;R. Misra
移动边缘计算中基于 DQN 的边缘缓存和动态服务迁移的能量延迟权衡
影响因子：
3.8
作者：
Chunlin Li;Yong Zhang;Xiang Gao;Youlong Luo
通讯作者：
Youlong Luo
DOI：
10.1016/j.jpdc.2022.01.020
发表时间：
2022-01
期刊：
J. Parallel Distributed Comput.
影响因子：
--
作者：
Chunlin Li;Yong Zhang;Youlong Luo
通讯作者：
Chunlin Li;Yong Zhang;Youlong Luo
影响因子：
4.1
作者：
Hou, Yanzhao;Wang, Chengrui;Wu, Xunchao
通讯作者：
Wu, Xunchao
地理分布式云中科学工作流程处理的容错调度和数据放置
影响因子：
3.5
作者：
Li, Chunlin;Liu, Jun;Luo, Youlong
通讯作者：
Luo, Youlong